[barmar]

Matmat's first example is not good. The tables where North showed its spades were where South doubled rather than overcalling 1NT.

But in the case of the second example, look at tables 1, 4, and 14. They're all board 31, with one human sitting South. At tables 1 and 14 East rebid 3♦, at table 4 it rebid 2♦.

Differences in play or defense are more common than differences in bidding, I think. I'm pretty sure some examples have been posted, I don't feel like trying to find them.

As for why I play with robots, I simply enjoy it, despite the flaws. Since I know the flaws, my expectations are relatively low, and it meets them. And when I make stupid mistakes, I don't have to worry about being embarassed, or worse getting some stupid comment from my partner about it.

[matmat]

1eyedjack, on 2011-July-19, 15:16, said:

I may have missed something in the thread or examples, but just because GIB is not guaranteed to make the same decision from the same identical premise in several successive occasions does not make the playing field unlevel. It may make it a bit more random, but in order for it to be unlevel you would have to demonstrate that against *you* it would be more inclined to take one action over another, contrasted with against another player, by reason of your identity rather than pure chance.

Strongly disagree with (and am somewhat amused by) this.
You're going from the initial position of "robots make the playing field level" to trying to redefine what "level playing field" means.

If we are going to claim that the GIBs lead to fairness then they should always act the same way given the same input information. As we know, GIBs are not deterministic and so we do expect to see some differences between its actions despite the same initial conditions. However, the variance that I think I am seeing indicates that either the decisions are really close and accurate simulations have a tough time telling the various actions apart or the simulations go through an inadequate number of hands and the error bars are too big. With how quickly (and often poorly) GIBs appear to be declaring these days (as opposed to what I seem to recall from the past) I would guess that the GIB is set to play way too fast and does not sample an adequate number of hands.

[matmat]

barmar, on 2011-July-19, 15:29, said:

Matmat's first example is not good. The tables where North showed its spades were where South doubled rather than overcalling 1NT.

You're right re my first one. I missed that and was about to post a correction just now (not that i think hiding ♠KQTxxx when p shows a strong NT is a good idea).

[xxhong]

I did some testing in the gib bidding long time ago. I still remember in one case, gib sometimes bid 7 and sometimes just stop at 6. I think it will take a lot of hands to make the final result converge, perhaps hundreds for many situations and thousands for others. So Gib has no solid way to make the final result converge in many situations because the sample size was around 10 (could be less nowadays). So, I kind of agree that the intrinsic design is rather bad for gib bidding and a lot of fluctuations should be expected in situations that requires a large sample size. Of course, bbo usually says that this kind of random distribution would even out in a long run.

One way to solve this problem is to do a large sample sized calculation when a simulation is needed and make a complete set of rules when the simulation is not needed. This is the way to minimize the simulations required. Also, the simulation can be run on a large parallel computer.
Now as it appears, gib makes too many simulations and can often override the system rules with simulation results that might not converge at all.

matmat, on 2011-July-19, 15:34, said:

Strongly disagree with (and am somewhat amused by) this.
You're going from the initial position of "robots make the playing field level" to trying to redefine what "level playing field" means.

If we are going to claim that the GIBs lead to fairness then they should always act the same way given the same input information. As we know, GIBs are not deterministic and so we do expect to see some differences between its actions despite the same initial conditions. However, the variance that I think I am seeing indicates that either the decisions are really close and accurate simulations have a tough time telling the various actions apart or the simulations go through an inadequate number of hands and the error bars are too big. With how quickly (and often poorly) GIBs appear to be declaring these days (as opposed to what I seem to recall from the past) I would guess that the GIB is set to play way too fast and does not sample an adequate number of hands.

[dustinst22]

jmcw, on 2011-July-19, 14:42, said:

Really!, I find that quite unbelievable. It is precisely the subtle spotting difference that may lead GIB to conclude an alternative defence, which is exactly how you preserve the level playing field.

For a robot to make a "mistake" it would have to produce a different result when the data imputs are reproduced and repeated exactly. Making bad plays or bids are not mistakes for a robot, but rather, the limit of expertise to which GIB has been programmed.

You might want to look again. 2 tables led the same spot card and the robot did something different. Here are the examples:

http://tinyurl.com/5ublkms

http://tinyurl.com/62xgrf7

Both tables played the exact same way (only the KC on trick 1 versus AC which shouldn't make a difference). The rest of the plays were identical. And please don't try to tell me the "subtle" lead of the KC made all the difference and is what made the GIB ruff in this instance(which is the incorrect lead, I might add, since GIB plays standard leads).

[fred]

When you play against GIBs in (for example) the Main Bridge Club, you should not expect identical bidding and play from the GIBs at every table because there can be at least 3 different versions of GIB in play. Furthermore, in at least 1 of these versions the user has control over GIB's "thinking time" and the value of this setting can impact the decisions that GIB makes.

The same is not supposed to be true in a robot tournament context. There all GIBs should behave the same way in the same circumstances. If you think you have seen a counter-example, it would be best if you sent a report to support@bridgebase.com so that we can investigate and fix the apparent bug.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

[dustinst22]

Thanks for the clarification, Fred. The above hands were in an ACBL tourney. I'll send to support.

[pigpenz]

matmat, on 2011-July-19, 11:43, said:

From the gripes i have seen on the forums and some of the results i have seen online, i think this is blatantly false.



This is, at least to me, unfortunate. I am glad that BBO is making sufficient profits off of this (I think) to keep the other services free, but the conclusion stated in this quote makes me sad.

I think on the playing field it shows how you stack up against others when it comes to YOUR technique playing the hand and bidding. Everyone is faced with the same problem, but I would prefer the choice of where to sit NSEW, not having the best hand at the table, and of playing either IMPS or MPpoints in the ACBL robot games.

[jmcw]

dustinst22, on 2011-July-19, 16:27, said:

You might want to look again. 2 tables led the same spot card and the robot did something different. Here are the examples:

http://tinyurl.com/5ublkms

http://tinyurl.com/62xgrf7

Both tables played the exact same way (only the KC on trick 1 versus AC which shouldn't make a difference). The rest of the plays were identical. And please don't try to tell me the "subtle" lead of the KC made all the difference and is what made the GIB ruff in this instance(which is the incorrect lead, I might add, since GIB plays standard leads).

I did not address my earlier post to you, nor would I presume to tell you anything. I don't know you!

Draw whatever conclusion you wish from the examples you attached.

However, it might be more helpful if you were to provide a possible explaination for the divergence in play. In my earlier post I offered one possible explaination, Fred has provide another (thankyou).

[pigpenz]

fred, on 2011-July-19, 21:39, said:

When you play against GIBs in (for example) the Main Bridge Club, you should not expect identical bidding and play from the GIBs at every table because there can be at least 3 different versions of GIB in play. Furthermore, in at least 1 of these versions the user has control over GIB's "thinking time" and the value of this setting can impact the decisions that GIB makes.

The same is not supposed to be true in a robot tournament context. There all GIBs should behave the same way in the same circumstances. If you think you have seen a counter-example, it would be best if you sent a report to support@bridgebase.com so that we can investigate and fix the apparent bug.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

generally when i go over the hands in the ACBL robot games the GIB is very consistent on leads and bidding following the constraints of the auction, if someone does something different then that has an effect....but still there are times I watch GIB and wonder why it does what it does.

takes this hold
96♥

K105♥

the 96♥ is in dummy and GIB has K105♥
declarer has opened 1♥........on almost ever hand when declarer played the 9♥ GIB played low, when the 6♥ was played GIB played the King....also GIB played the King♥ in several instances when declarer played the second finesse....but what caught my eye was the difference in the play when the 9 or 6 ♥ was lead initially.

[Bbradley62]

fred, on 2011-July-19, 21:39, said:

When you play against GIBs in (for example) the Main Bridge Club, you should not expect identical bidding and play from the GIBs at every table because there can be at least 3 different versions of GIB in play. Furthermore, in at least 1 of these versions the user has control over GIB's "thinking time" and the value of this setting can impact the decisions that GIB makes.

The same is not supposed to be true in a robot tournament context. There all GIBs should behave the same way in the same circumstances. If you think you have seen a counter-example, it would be best if you sent a report to support@bridgebase.com so that we can investigate and fix the apparent bug.

Fred Gitelman
Bridge Base Inc.
www.bridgebase.com

Pigpenz reported this in a GIB tournament, where some Norths responded 2D to a 1S opener, but some responded 1NT. http://online.bridge...sername=pigpenz